I was scrolling through twitter and found this\nThe image shows an annotated page of kuzushiji and claims that a model \"KuroNet\" can recognize the characters in an entire page in about 2 seconds.\nUnfortunately, I couln't find any code/paper about KuroNet.
\n\nDoing some more twitter \"research\" I came across this which reads:\n> This will be out on arxiv next month with the code on github. NekoNet (Neural End-to-end Kuzushiji OCR) is a simple U-Net model for Kuzushiji recognition. This model is not SOTA (KuroNet is), but it’s easy to understand and implement so it’s baseline model.
\n\nI checked on github and arxiv and couldn't find the code or paper but I did find this, a page that has the download of the paper, though, behind a paywall
\n\nIf anyone finds access to the NekoNet code or paper, please share it with the rest of us!
\n\nAlso, if @tkasasagi could provide it herself, that would also be very much appreciated!
\n\nEdit: here is another helpful resource, it allows you to go through photos of japanese text, highlight a character and it will recognize it for you!
", "messages": [ { "id": 586086, "postDate": "2019-07-28T15:17:03.557Z", "content": "I was scrolling through twitter and found this\nThe image shows an annotated page of kuzushiji and claims that a model \"KuroNet\" can recognize the characters in an entire page in about 2 seconds.\nUnfortunately, I couln't find any code/paper about KuroNet.
\n\nDoing some more twitter \"research\" I came across this which reads:\n> This will be out on arxiv next month with the code on github. NekoNet (Neural End-to-end Kuzushiji OCR) is a simple U-Net model for Kuzushiji recognition. This model is not SOTA (KuroNet is), but it’s easy to understand and implement so it’s baseline model.
\n\nI checked on github and arxiv and couldn't find the code or paper but I did find this, a page that has the download of the paper, though, behind a paywall
\n\nIf anyone finds access to the NekoNet code or paper, please share it with the rest of us!
\n\nAlso, if @tkasasagi could provide it herself, that would also be very much appreciated!
\n\nEdit: here is another helpful resource, it allows you to go through photos of japanese text, highlight a character and it will recognize it for you!
", "rawMarkdown": "I was scrolling through twitter and found [this](https://twitter.com/tkasasagi/status/1155357957623668737)\nThe image shows an annotated page of kuzushiji and claims that a model \"KuroNet\" can recognize the characters in an entire page in about 2 seconds.\nUnfortunately, I couln't find any code/paper about KuroNet.\n\nDoing some more twitter \"research\" I came across [this](https://twitter.com/tkasasagi/status/1133269028162588673?lang=en) which reads:\n> This will be out on arxiv next month with the code on github. NekoNet (Neural End-to-end Kuzushiji OCR) is a simple U-Net model for Kuzushiji recognition. This model is not SOTA (KuroNet is), but it’s easy to understand and implement so it’s baseline model.\n\nI checked on github and arxiv and couldn't find the code or paper but I did find [this](https://ipsj.ixsq.nii.ac.jp/ej/?action=pages_view_main&active_action=repository_view_main_item_detail&item_id=192436&item_no=1&page_id=13&block_id=8), a page that has the download of the paper, though, behind a paywall\n\nIf anyone finds access to the NekoNet code or paper, please share it with the rest of us!\n\nAlso, if @tkasasagi could provide it herself, that would also be very much appreciated!\n\nEdit: [here](http://codh.rois.ac.jp/char-shape/app/icv-kuzushiji/?manifest=http://codh.rois.ac.jp/char-shape/book/100241706/manifest.json&pos=70&lang=en) is another helpful resource, it allows you to go through photos of japanese text, highlight a character and it will recognize it for you!", "votes": 12 } ], "comments": [], "raw_markdown_by_id": { "586086": "I was scrolling through twitter and found [this](https://twitter.com/tkasasagi/status/1155357957623668737)\nThe image shows an annotated page of kuzushiji and claims that a model \"KuroNet\" can recognize the characters in an entire page in about 2 seconds.\nUnfortunately, I couln't find any code/paper about KuroNet.\n\nDoing some more twitter \"research\" I came across [this](https://twitter.com/tkasasagi/status/1133269028162588673?lang=en) which reads:\n> This will be out on arxiv next month with the code on github. NekoNet (Neural End-to-end Kuzushiji OCR) is a simple U-Net model for Kuzushiji recognition. This model is not SOTA (KuroNet is), but it’s easy to understand and implement so it’s baseline model.\n\nI checked on github and arxiv and couldn't find the code or paper but I did find [this](https://ipsj.ixsq.nii.ac.jp/ej/?action=pages_view_main&active_action=repository_view_main_item_detail&item_id=192436&item_no=1&page_id=13&block_id=8), a page that has the download of the paper, though, behind a paywall\n\nIf anyone finds access to the NekoNet code or paper, please share it with the rest of us!\n\nAlso, if @tkasasagi could provide it herself, that would also be very much appreciated!\n\nEdit: [here](http://codh.rois.ac.jp/char-shape/app/icv-kuzushiji/?manifest=http://codh.rois.ac.jp/char-shape/book/100241706/manifest.json&pos=70&lang=en) is another helpful resource, it allows you to go through photos of japanese text, highlight a character and it will recognize it for you!" } }